COCO
Benchmarks
TaskDatasetSOTA resultTrendResults
COCO standard (transductive)
71Win Rate
3
COCO inductive rev.
84Win-rate
3
COCO (inductive)
70Win Rate
3
COCO (val)
82.47Surrogate Acc (Initial Constitution)
3
COCO 2017 (val)
44.6AP
3
COCO Caption
43.2BLEU@4
3
COCO-OOC
97.2Accuracy
3
COCO (val)
2.11FID
3
COCO
10.93PSNR
3
MS-COCO (test val)
0.29RD
3
COCO 2014 (val)
8.3rFID
3
COCO
57.8CIDEr-D
3
COCO (test)
100.72CIDEr (Avg 1-2)
3
COCO-OCC (val)
31.7AP
3
COCO (test)
35.76BRI
3
COCO 5,000 samples 2017 (val)
20.61FID
3
COCO 5,000 samples 2017 (val)
20.27FID
3
COCO human evaluation
4.9Fluency
3
Hal-COCO
88.3Top-1 Accuracy
3
COCO
93Recall
3
COCO K (test)
152.1CIDEr
3
COCO Capt.
126CIDEr
3
COCO Caption
77BLEU-1
3
COCO multiple objects (val)
5.89FID
3
COCO single object (val)
3.78FID
3